Threads and Critical Sections Thomas Plagemann Slides from Otto J. Anshus, Tore Larsen (University...

Threads and Critical Sections

Thomas PlagemannSlides from Otto J. Anshus, Tore Larsen (University of Tromsø),

Kai Li (Princeton University)

Overview

• Intro to threads• Concurrency• Race conditions & critical regions• Too much milk problem• -> mutual exclusion …..

Thread and Address Space

• Thread– A sequential execution stream within a

process (also called lightweight process)• Address space

– All the state needed to run a program – Provide illusion that program is running on

its own machine (protection)– There can be more than one thread per

address space

Typical Thread API

• Creation– Fork, Join

• Mutual exclusion– Acquire (lock), Release (unlock)

• Condition variables– Wait, Signal, Broadcast

• Alert– Alert, AlertWait, TestAlert

• Difficult to use

• Not good: Combines specification of concurrency (Fork) with synchronization (Join)

Fork/Join

P1:

….

FORK T1;

….

JOIN T1;

….

T1:

….

END;

P1 must WAIT until T1 finishes

Executes concurrently

P1:

….

FORK T1;

FORK T2;

FORK T3;

….

JOIN;

….

T1:

….

END;

T2:

….

END;T3:

….

END;JOIN whom??

Figure 2-14. Some of the Pthreads function calls.

POSIX Threads (1)

Tanenbaum, Modern Operating Systems 3 e, (c) 2008 Prentice-Hall, Inc. All rights reserved. 0-13-6006639

Figure 2-15. An example program using threads.

POSIX Threads (2)

Tanenbaum, Modern Operating Systems 3 e, (c) 2008 Prentice-Hall, Inc. All rights reserved. 0-13-6006639

. . .

Figure 2-15. An example program using threads.

POSIX Threads (3)

Tanenbaum, Modern Operating Systems 3 e, (c) 2008 Prentice-Hall, Inc. All rights reserved. 0-13-6006639

. . .

User vs. Kernel-Level Threads

• Question– What is the difference between user-level

and kernel-level threads?• Discussions

– When a user-level thread is blocked on an I/O event, the whole process is blocked

– A context switch of kernel-threads is expensive

– A smart scheduler (two-level) can avoid both drawbacks

User vs. Kernel Threads

KERNEL

Thread Package

ThreadsThreads

Thread PackageProcess “Package”

Recall last week: PCB resp. PT

• Which information has to be stored/saved for a process?

Thread Control Block

• Shared information– Processor info: parent process, time, etc– Memory: segments, page table, and stats,

etc– I/O and file: comm ports, directories and file

descriptors, etc• Private state

– State (ready, running and blocked)– Registers– Program counter– Execution stack

System Stack for Kernel Threads

• Each kernel thread has– a user stack– a private kernel stack

• Pros– concurrent accesses to system services– works on a multiprocessor

• Cons– More memory

• Each kernel thread has– a user stack– a shared kernel stack with other threads in the same

address space• Pros

– less memory• Cons

– serial access to system services

Typical for all shared resources

System Stack for Kernel Threads in Linux

http://jon.oberheide.org/blog/2010/11/29/exploiting-stack-overflows-in-the-linux-kernel/

Concurrency and Threads

• I/O devices– Overlap I/Os with I/Os and computation (modern OS

approach)• Human users

– Doing multiple things to the machine: Web browser• Distributed systems

– Client/server computing: NFS file server• Multiprocessors and multi cores

– Multiple CPUs sharing the same memory: parallel program

Why is concurrency tricky

• We use threads or processes as powerful concept to keep threads of control apart, looking at sequential processes instead of arbitrarily interleaved thread of execution

• Yet, in reality interleaved, any process may be interrupted at any time

• Subtle, hard-to-identify, programming errors including race-conditions are introduced

17

Example: NASA Pathfinder spacecraft

• Total system resets in Mars Pathfinder• An overrun of a data collection task

a priority inversion in mutex semaphore a failure of a communication task a system reset.

• Took 18 hours to reproduce the failure in a lab replica the problem became obvious and a fix was installed

• Errors rooted in the interaction of multiple concurrent operations/threads and are based on timing dependencies.

• Easy to identify the errors and fix them once the failing sequences are reproduced (or observed).

Source: Yann–Hang Lee, Gerald Gannod, and Karam Chatha, Arizona State UniversityW. Eric Wong, University of Texas

Priority Inversion on Mars Pathfinder

Source: http://www.cse.nd.edu/~cpoellab/teaching/cse40463/slides6.pdf

19

Temporal Dependence

• Predicting and controlling timing and responses are based on event occurrences

• Timing relationship: (can you guarantee it?)– Predictable actions in response to external stimuli

• if event E1 occurs at time t1, will an action A1 be triggered at time t2

– Deadline (absolute or relative), and jitter• Program execution

• If event E1 occurs at time t1+, will the same action A1 be triggered at time t2 + ?

• Will the program execution be identical ?• Should this case be tested ?

Source: Yann–Hang Lee, Gerald Gannod, and Karam Chatha, Arizona State UniversityW. Eric Wong, University of Texas

Concurrency: Double buffering

Put (t,g)

/* Copy */ t := s;

Input sequence f

Output sequence g

Get (s,f) s

t

Get(s,f);

Repeat

Copy;

cobegin

Put(t,g);

Get(s,f);

coend;

until completed;

/* Fill s and empty t concurrently */

• Put and Get are disjunct

• … but not with regards to Copy!

(Threads)

Specifies concurrent execution

Concurrency: Time Dependent Errors

Repeat

cobegin

Copy;

Put(t,g);

Get(s,f);

coend;

until completed;

Repeat

Copy;

cobegin

Put(t,g);

Get(s,f);

coend;

until completed;

• C-P-G

• C-G-P

• P-C-G

• P-G-C

• G-C-P

• G-P-C

The rightmost (incorrect) solution can be executed in 6 ways:

Interleaving!

In the correct solution we solved the problem of sharing of the buffers between Copy and Put/Get by designing an algorithm avoiding problems

Mini assignment: are both solutions correct? What can happen?

22

Race Conditions• Necessary conditions:

– Concurrent operations– At least one is “update”– No mechanism to guarantee any specific order

3 operations – A, B, & C• A race condition occurs when two threads manipulate a

shared data structure simultaneously without synchronization and the result depends on who runs precisely when.

• Race conditions are common errors in multi-threaded programs; Since they are timing-dependent, they are notoriously hard to catch during testing

• Possible consequences:– inconsistent data– unexpected (non-deterministic) execution sequence (order of

actions)

A CB

C AB

B CA

Source: Yann–Hang Lee, Gerald Gannod, and Karam Chatha, Arizona State UniversityW. Eric Wong, University of Texas

Critical Region

• Mutual exclusion• The part of the program where the shared memory (or

something else) is accessed is called a critical section• Four conditions for a good solution for mutual exclusion:

– Not two processes simultaneously in their critical regions– No assumptions may be made about speed and number of CPUs– No process running outside its critical region may block another

process– No process should have to wait forever to enter its critical region

Figure 2-22. Mutual exclusion using critical regions.

Critical Regions (2)

Tanenbaum, Modern Operating Systems 3 e, (c) 2008 Prentice-Hall, Inc. All rights reserved. 0-13-6006639

“Too Much Milk” Problem

• Don’t buy too much milk• Any person can be distracted at any

point

Person A Person B

Look in fridge: out of milkLeave for Rema1000Arrive at Rema1000Buy milkArrive home

Look in fridge: out of milkLeave for Rema1000Arrive at Rema1000Buy milkArrive home

A Possible Solution?

if ( noMilk ) { if (noNote) { leave note; buy milk; remove note; }}

if ( noMilk ) { if (noNote) { leave note; buy milk; remove note; }}

A: B:

A Possible Solution?

if ( noMilk ) { if (noNote) { leave note; buy milk; remove note; }}

Ping!!!: and B starts executing until finished, and then A starts again

if ( noMilk ) { if (noNote) { leave note; buy milk; remove note; }}

A: B:

(But B will “see” A by the fridge?: That is what we are trying to achieve.)

And both A and B buys milk.

The ENTRY is flawed

Another Possible Solution?

Thread A

leave noteAif (noNoteB) { if (noMilk) { buy milk }}remove noteA

Thread B

leave noteBif (noNoteA) { if (noMilk) { buy milk }}remove noteB

Another Possible Solution?

Thread A

leave noteAif (noNoteB) { if (noMilk) { buy milk }}remove noteA

Thread B

leave noteBif (noNoteA) { if (noMilk) { buy milk }}remove noteB

Ping!! And B starts

Ping!! And A starts

“Milk starvation” possible, but perhaps not a problem in practice!

WHY?

Yet Another Possible Solution?

Thread A

leave noteAwhile (noteB) do nothing;if (noMilk) buy milk;remove noteA

Thread B

leave noteBif (noNoteA) { if (noMilk) { buy milk }}remove noteB

Yet Another Possible Solution?

• Safe to buy• If the other buys, quit

Thread A

leave noteAwhile (noteB) do nothing;if (noMilk) buy milk;remove noteA

Thread B

leave noteBif (noNoteA) { if (noMilk) { buy milk }}remove noteB

• Not symmetric solution

• Busy wait!

Remarks

• The last solution works, but– Life is too complicated– A’s code is different from B’s– Busy waiting is a waste

• Peterson’s solution is also complex• What we want is:

Acquire(lock);if (noMilk) buy milk;Release(lock);

Critical section a.k.a. Critical region a.k.a. Mutual Exclusion (Mutex)

Entry and Exit Protocols

ENTRY;

<Critical region>;

EXIT;

Entry and Exit Protocols

ENTRY;

<Critical region>;

EXIT;

Threads blocked waiting to get access

Will release no. 1 in queue so it can enter CR

All threads must conform to the structure:

ENTRY;

<use resources reserved>

EXIT;

What will happen if they don’t?

Characteristics of a realistic solution for Mutual Exclusion

• Mutex: Only one process can be inside a critical region

• Non-preemptive scheduling of the resource: A thread having the resource must release it after a finite time

• No one waits forever: When the resource is requested by several threads concurrently, it must be given to one of them after a finite time

• No busy wait (?)• Processes outside of critical section should not

block other processes• No assumption about relative speeds of each

thread (time independence)• Works for multiprocessors

Summary

• Concurrency• Threads first intro• Too much milk problem

→ mutual execution!• Entry & exit

• Tomorrow: mutual exclusion with HW support